Re: [sl4] Convergence of Expected Utilities with Algorithmic Probability Distributions - uh?

From: Joshua Fox (
Date: Sun Dec 07 2008 - 08:10:45 MST

Since we're discussing the paper here, I hope that some questions about it
are not out of place.

* A broad question, not a technical criticism of the proof, but rather a
request for an intuitive understanding:
I'm wondering why p cannot nose-dive as fast as U skyrockets. I see that
you bound both U and p from below with computable functions. Of course,
that's not too tight a bound, but it raises the question: Why can't p go
down as much as U goes up, so that ultimately the series of their products

 * A minor notational question: In the proof of lemma 1.
F(x) = 1 + f(x) + max {B(x) − f(x) : x ∈ N}
The x in the curly-braces is quantified and the x outside the braces is
not; does that make these in fact separate variables?


On Thu, Dec 4, 2008 at 3:58 PM, Peter de Blanc <>wrote:

> Joshua Fox wrote:
>> In fact, De Blanc says (to simplify greatly), your utility function must
>> be bounded from above.
> From below, too.
> The paper is built on some powerful (big) assumptions:
> 1. You consider all computer programs as possible descriptions of the
> universe.
> 2. You have a utility function which is computably determined by your
> perceptions.
> I think (1) is fine (but Eli has objected to it), but (2) seems dubious to
> me.
> - Peter de Blanc
> __________________________________________________
> D O T E A S Y - "Join the web hosting revolution!"

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:03 MDT